FIC WAN frequent itemset clustering of web articles by analyzing the article neighborhood

Abstract

Document clustering is a process of organizing text data into clusters where a cluster usually represents a group of topic related documents. Most effective text clustering approaches are based on frequent itemsets. A popular algorithm that uses this approach is FIHC (Frequent Itemset-based Hierarchical Clustering). In recent years, many modifications have been made to this algorithm. In this paper we focus on clustering web articles which represent a special type of text data. They contain hyperlinks through which they are linked with other articles on the web. We propose a FICWAN algorithm which is a modification of FIHC. FICWAN is especially suited for web data. We show that by considering the neighborhood of a web article and its HTML tags and CSS we are able to significantly improve the quality of created clusters. We experimented with our approach on several corpuses and the results clearly outperformed FIHC.